Topic
Approval gates that pause agent actions on real systems until a person authorises them.
Gerard Smyth credits meeting automation with 10 to 20 hours a week per advisor and AI summaries with five to 10 more, but the $1bn budget behind it comes with no advisor count and no revenue line to check it against.
Reality
Checkmarx's lies-in-the-loop work shows the dialog a developer approves is rendered from the untrusted context the agent just read, metadata line included, which puts the last safeguard on the wrong side of the trust boundary.
Reality
Two additions push a coding agent past editing source, into watching a running application and auditing a repository. Both rely on access that two documented flaws have already abused.
Reality
A dev.to post argues that agents chaining tool calls fail as an architecture problem rather than a prompting one. Its confirmation gate on write tools holds even when the model's own confidence number is wrong.
Reality
The open-source framework ships durable execution, sandboxing, approvals, subagents and evals in public preview. If the bet lands, your in-house harness is now maintenance.
Reality